Prediction of SPI drought index using support vector and multiple linear regressions

Drought is a natural phenomenon, which has a complex mechanism as a result of interactions of meteorological parameters and usually occurs in all climates. So, predicting drought indices and their chronological evaluation is an effective way for the drought management and adaptation with its consequences. In the current research, prediction of drought indices are carried out for Tabriz synoptic station, using support vector regression, multiple linear regression and standard precipitation index (SPI) for the time period of 1979 to 2012. In this regard, for predicting SPI indices in the periods of 3, 6, 9, 12, 24 and 48 months, six different input combinations including the antecedent correspondent values of the mentioned index have been utilized. The results of statistical analysis showed that both methods had significant accuracy. Nonetheless, the support vector regressions for predicting SPI-6, SPI-9 and SPI-24 had better performances, regarding the root mean squared errors of 0.4985, 0.4340 and 0.2427, respectively. However, the multiple linear regressions showed lower relative errors, for predicting SPI-3, SPI-12 and SPI-48. Meanwhile, it can be concluded that both examined methods including support vector and multiple linear regressions had acceptable predictions of drought index and can be used with an admissible confidentiality for the management of drought consequences.